Assessment of Various Bias Correction Methods on

Precipitation of Regional Climate Model and Future

Projection
Gunavathi S
SASTRA University School of Civil Engineering: Shanmugha Arts Science Technology and Research
Academy School of Civil Engineering
Selvasidhu R (  selvakumar@civil.sastra.edu )
SASTRA University School of Civil Engineering: Shanmugha Arts Science Technology and Research
Academy School of Civil Engineering https://orcid.org/0000-0002-1635-8977

Research Article

Keywords: CORDEX, Precipitation, Future projection, Bias correction, Power transformation method,
Thanjavur district

Posted Date: April 7th, 2021

DOI: https://doi.org/10.21203/rs.3.rs-339080/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
1 Corresponding Author: Selvakumar Radhakrishnan

1
3 ASSESSMENT OF VARIOUS BIAS CORRECTION METHODS ON PRECIPITATION OF REGIONAL
4 CLIMATE MODEL AND FUTURE PROJECTION

5 Gunavathi S1., Selvakumar Radhakrishnan2

1, 2
6 School of Civil Engineering, SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India.
7 Email id2: selvakumar@civil.sastra.edu

9 Abstract:

10 The application of regional climate model simulations (RCMs) in climate change impact studies is
11 challengeable due to the risk of possible biases. Some sort of correction needs to be done prior to the application of
12 RCM simulations. This study attempts to assess the performance of a simple (linear scaling and Delta Change
13 method) and complex correction technique (Local intensity scaling, Power transformation and Distribution
14 mapping) on CORDEX(Coordinated Regional Climate Downscaling Experiment)simulated precipitation series for
15 the Thanjavur district. The performance at annual resolution is evaluated using various statistical parameters such as
16 Correlation, Root Mean Square Error and Bias against the observed precipitation data. The raw RCM estimates were
17 improved significantly after the bias correction with all methods. However, Power transformation exhibits good
18 agreement with the observed data at the district level than other methods because it corrects both the mean and
19 variance. The future climate was projected from 2021 to 2100 for RCP 4.5 and RCP 8.5 scenarios. The temporal
20 distribution of future precipitation clearly shows that most of the years will receive heavy precipitation; rather, some
21 years will receive low and average precipitation. The spatial distribution pattern indicates that the northeast
22 monsoon will dominate over all the ranges and places. This study has provided clear information on future
23 precipitation to the environmentalist, urban planners, and policymakers to take appropriate mitigation measures
24 towards agriculture and disaster management. Rainwater harvesting, recharging the aquifers, afforestation, and
25 redirecting the excess amount of water to the river through proper channels is the plausible actions suggested
26 overcoming excessive precipitation in the future.
27
28
29
30 Keywords: CORDEX, Precipitation, Future projection, Bias correction, Power transformation method, Thanjavur
31 district
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46

2
 47 1.0 Introduction:
48 Climate change, a complex phenomenon and critical issue being faced by the earth, has posed immense
49 threats to the ecological and human environment. The Intergovernmental Panel on Climate Change (IPCC)
50 reported that in the past few decades, global temperature, precipitation patterns and the occurrence of disasters
51 such as cyclones, droughts, floods and heatwaves had risen notably in terms of both frequency and severity
52 (IPCC 2013; Prusty et al. 2018). Though climate change is a long-term event, its impact may be imperceptibly
53 gradual and steady. Therefore, evaluating its effects and susceptibility to adaptation requires better knowledge
54 about the future climate. The research community worldwide is currently using coarse and high-resolution
55 climate models for climate predictions and assessment (Giorgi et al. 2018). In general, the course resolution
56 Global Climate Models (GCMs) has shortcomings in capturing regional orographic characteristics. In contrast,
57 high-resolution Regional Climate Models (RCMs) reflect improved orography and generate more accurate
58 climate projection. Hence commonly used to evaluate the past, present and future climates across the world
59 (Kumar et al. 2013).
60
61 Accordingly, the Coordinated Regional Climate Downscaling Experiment (CORDEX), a coordinated
62 initiative involving several countries worldwide, produces various regional and local climate simulations
63 (Giorgi et al. 2009). Even though RCM simulations are driven using multiple GCMs with the improved
64 horizontal resolution, it yields errors. Biases are due to the uncertainty in physical parameterizations, lateral
65 boundary conditions, initial condition limitation, numerical model imperfectness, and so on (Giorgi et al. 1999;
66 Christensen et al. 2008; Rauscher et al. 2010; Hall et al. 2014). In consequence, reduces the reliability and
67 increases the uncertainty in using the RCM simulations directly as input data for climate change studies
68 (Noguer et al. 1998). However, earlier studies have concluded that the climate model outputs shall be refined by
69 applying statistical corrections (Veijalainen et al. 2010; Dosio et al. 2011; Hanel et al. 2011; Monhart et al.
70 2018; Pontoppidan et al. 2018). Researchers use multiple bias correction methods for reducing model errors and
71 also to downscale the GCM data. The techniques vary from basic scaling to very complex methods like weather
72 generators, probability mapping, etc. Overall, bias correction methods use the transformation algorithm for
73 correcting the RCMs output by identifying the bias between observed and simulated data. The derived bias
74 correction algorithm and parameterization are also used to correct the RCM projection scenarios (Chen et al.
75 2011b; Johnson et al. 2011).
76
77 In general, corrections are made by correcting the mean, wet day frequencies and percentile on the
78 simulated data with reference to the observed data (Gudmundsson et al. 2012; Teutschbein et al. 2012, 2013;
79 Chen et al. 2013). Based on the mean, standard deviation, and coefficient of variation of the observed data,
80 Terink et al. (2009) adjusted the daily RCM-simulated precipitation and temperature data for the Rhine basin.
81 The distribution mapping method was used to adjust the RCM simulated daily precipitation dataset over Europe
82 and found that this method worked relatively well under normal and extreme conditions (Piani et al. 2010).
83 Themeßl et al. (2012) found that quantile mapping and local intensity scaling (LOCI) methods effectively
84 correct the Alpine region's daily precipitation simulation by analyzing seven bias correction techniques.
85 Similarly, Bennett et al. (2011) also used quantile mapping to correct annual and seasonal RCM rainfall
86 deviations in Australia and noted an improved spatial distribution after correction. To reduce the bias of RCM-
87 simulated precipitation for seven catchments across the United Kingdom, Lafon et al. (2013) assessed linear,
88 nonlinear, gamma and quantile mapping based on empirical distribution methods. The RCM simulated
89 precipitation dataset has also been corrected using distribution, parametric and nonparametric transformations
90 for 83 stations in Norway by Gudmundsson et al. (2012). N'Tcha M'Po et al. (2016) found that the empirical-
91 based quantile mapping approach works better than Gamma-based quantile mapping to correct extreme
92 precipitation. Three distinct processes, such as linear scaling, regression and empirical quantile mapping, were
93 compared to select the most effective method over the Northwest Himalayas using four statistical measurements
94 by Devi et al. (2021). Thus, most research is conducted to identify the best method that shows good agreement
95 with observed data. Accordingly, in the present study, linear scaling, local intensity scaling, power
96 transformation, distribution mapping, and delta-change methods are evaluated on REMO 2009 RCM. The
97 projection of precipitation using the best bias correction method RCP 4.5 and RCP 8.5 scenarios is carried out.
98
99 2.0 Study area:
100 Thanjavur district located between 9º 50' and 11º 25' North and 78º 45' and 79º 25' East with the total
101 geographical area of 3,602.86 sq. km has been studied. The region possesses three rainy seasons such as

3
102 summer rain (March-May), South-West monsoon (June-September) and North-East monsoon (October-Early
103 January). Amongst, the North-East monsoon (545.7 mm of normal rainfall) and the South-West monsoon (342
104 mm of normal rainfall) plays a significant role in feeding the river Cauvery, the primary source of irrigation of
105 the study area. The area is unique for its agricultural activities from time immemorial and is renowned as
106 the granary of South India. Besides the river Cauvery, the area has been crisscrossed by a network of irrigation
107 channels. Hence, this coastal district is flourishing in paddy fields, coconut and mango groves, plantain trees
108 and other vegetation. However, in recent times, agriculture seems to get destabilized due to uncertain climatic
109 conditions. The study area map shows 14 blocks with 17 well-distributed rain gauge stations around the district
110 (Figure1).

111 3.0 Materials and Methods:

112 The four steps methodology has adopted in this study (Figure 2).

113 3.1 Collection of observed and model data:

114 In this study, the daily precipitation data were collected for 30 years (1976 – 2005) from the Indian
115 Meteorological Department (IMD). The obtained data was found with some gaps. Hence observed proxy data
116 (APHRODITE - Asian Precipitation – Highly –Resolved Observational Data Integration Towards Evaluation)
117 (http://aphrodite.st.hirosaki-u.ac.jp/) and reanalysis data (ERA - INTERIM)
118 (https://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc/) were collected, and the annual mean compared
119 against the observed data. In that, APHRODITE data has shown a high correlation of 0.73 with the observed data.
120 Thus, observed IMD data was primarily used in the present study, while the data gaps were filled with
121 APHRODITE data.

122 CORDEX is a global programme for localized climate change scenarios. CORDEX South Asia domain
123 experiment consists of eleven distinct suites, with different RCMs driven by various initial and boundary
124 forcing GCMs. Within the context of CORDEX, the Max Planck Institute for Meteorology is extending its regional
125 climate model REMO to many regions of the planet. This study employed the daily precipitation datasets of the
126 REMO 2009 simulations driven by the MPI-MPI-ESM-LR driving model of CORDEX-SOUTH ASIA domain
127 (WAS-44i; ~50 km horizontal resolution). The data is available in the Earth Systems Grid Federation (ESGF) under
128 the CORDEX project (https://esgf-index1.ceda.ac.uk/search/esgf-ceda/).

129 3.2 Selection of bias correction techniques:

130 As stated earlier, datasets were collected on daily observed data (IMD) and RCM Simulated precipitation
131 for a control period of 30 years (1976–2005) for bias correction. Hereafter, the RCM simulated data for the
132 historical period will be called 'control data' and for the future period as 'scenario data'. Amongst various bias
133 correction methods, the following were used to adjust the control data in the present study (1) linear scaling, (2)
134 local intensity scaling, (3) power transformation, (4) distribution mapping and (5) delta-change approach.

135 3.2.1 Linear scaling of precipitation (LS):

136 LS is the most straightforward bias correction technique employed in several studies (Ines et al. 2006;
137 Teutschbein et al. 2012; Shrestha et al. 2015). It adjusts the RCM mean value with a perfect agreement with the
138 observation data. The control and scenario precipitation are then adjusted based on the ratio between the long-term
139 monthly mean observed and control/scenario data using equations (1) and (2), respectively. However, this approach
140 can correctly adjust the climatic factors only when the monthly mean values are included.
141
éµ (P (d ) ) ù
142 *
Pcontrol (d ) = Pcontrol (d ) × ê m observed ú (1)
êë µm ( Pcontrol (d ) ) úû

éµ (P (d ) ) ù
143 *
Pscenario (d ) = Pscenario (d ) × ê m observed ú (2)
êë µm ( Pcontrol (d ) ) úû

4
144 Where, P = precipitation; (d) = daily time series; µ = mean and P* = final bias corrected.

145 3.2.2 Local intensity scaling (LOCI) of precipitation:

146 The LOCI method introduced by Schmidli et al. (2006) extends the linear scaling method a step forward.
147 Added to the mean, it also adjusts wet-day frequencies and wet-day intensities of precipitation. The precipitation
148 intensity threshold (Pth, control) for every month is initially confirmed. Then, the number of wet days in control data
149 that exceeds the threshold will be adjusted based on the number of days the observed precipitation was determined.
150 The number of precipitation events for control and scenario run is corrected by applying the calibrated RCM
151 precipitation threshold (Pth, control) using Equations (3) and (4), respectively. This approach virtually eliminates the
152 drizzle effect because excessive drizzly days are frequently added to the RCM outputs.
153
ì0, if Pcontrol ( d ) < Pth ,control
154 *1
Pcontrol ( d ) = ïí (3)
ïîPcontrol ( d ) , otherwise

ì0, if Pscenario ( d ) < Pth ,control

155 *1
Pscenario ( d ) = ïí (4)
ïîPscenario ( d ) , otherwise

156 A scaling factor s is then calculated using equation (5) to confirm that the mean of corrected precipitation is equal to
157 observed data.

158 s=
(
µm Pobserved ( d ) Pobserved ( d ) > 0 mm ) (5)
µm ( Pcontrol ( d ) Pcontrol ( d ) > Pth ,control ) - Pth.control

159 Finally, both control and scenario precipitations are corrected using equation (6) and (7), respectively.
* *1
160 Pcontrol ( d ) = Pcontrol (d ) × s (6)

* *1
161 Pscenario ( d ) = Pscenario (d ) × s (7)

162 Where, th = threshold and P*1 = intermediate step in bias correction.

163 3.2.3 Power transformation of precipitation (PT):

164 PT corrects both the monthly mean as well as the variance. It uses an exponential correcting factor aPb
165 (Mendez et al. 2020; Kim et al. 2020). Parameter 'b' is measured monthly (bm) using the distribution-free method
166 with a three-month window. Initially, 'b' is determined by equalizing the Coefficient of Variation (CV) of corrected
167 RCM precipitation (Pb) and observed precipitation (Pobserved) for every month (m) using the root-finding algorithm
168 (Brent 1971). Then 'bm' is calculated using equation eight and 'CVm' using equation nine. Equation 10 & 11 were
169 used for equalizing the datasets.
170
171 bm
(
f ( bm ) = 0 = CVm ( Pobserved ( d ) ) - CVm Pcontrol (d ) ) (8)

172
s m ( Pobserved ( d ) ) s m ( Pcontrol ( d ) )
b m

-
µm ( Pobserved ( d ) ) µm ( Pcontrol
bm
( d ))
173 (9)

*1 bm
174 Pcontrol ( d ) = Pcontrol (d ) (10)

5
 *1 bm
175 Pscenario ( d ) = Pscenario (d ) (11)

176 Afterwards, 'PT' equalizes the observed precipitation (Pobserved) with the intermediate series (P*1control) using
177 the LS method. Finally, the corrected control and scenario precipitation datasets were derived using equations 12
178 and 13, respectively.
éµ (P ( d )) ù
179 *
Pcontrol *1
( d ) = Pcontrol ( d ) × ê m observed ú (12)
(
êë µm Pcontrol
*1
)
( d ) úû

éµ (P ( d )) ù
180 *
Pscenario *1
( d ) = Pscenario ( d ) × ê m observed ú (13)
(
êë µm Pcontrol
*1
( d ) úû)
181 3.2.4 Distribution mapping of precipitation (DM):

182 The DM method is applied to correct mean, standard deviation (SD), and quantiles by equalizing the
183 distribution functions of both the RCM outputs and the observed data. The method assumes that the RCM-simulated
184 and observed precipitation follows a particular frequency of distribution, in turn, may cause biases (Themeßl et al.,
185 2012). Accordingly, Gamma distribution was used for effective precipitation distribution.

-c
1
186 fg ( c a , b ) = c a -1
× a × e b ; c ³ 0; a, b > 0 (14)
b × G (a )

187 Where Γ(.) is the Gamma function, α is the shape parameter, and β is the scale parameter. Before the DM
188 method, the LOCI method is applied to determine the wet days using the specific threshold. Subsequently, RCM
189 outputs were corrected in terms of the Gamma cumulative distribution function (Fγ) and its inverse function (F-1γ) as
190 follows:
191 *
Pcontrol ( ( )
( d ) = Fg-1 Fg Pcontrol ( d ) a control ,m, bcontrol ,m a observed ,m, bobserved ,m ) (15)

192 *
Pscenario ( ( )
( d ) = Fg-1 Fg Pscenario ( d ) a control ,m, bcontrol ,m a observed ,m, bobserved ,m ) (16)

193 3.2.5 Delta Change Method (DC):

194
195 DC method is comparatively simple and widely used (Middelkoop et al. 2001; Raty et al. 2014). The
196 difference between the mean of the scenario and the control data is calculated. This change/delta is added with the
197 monthly mean of observed data to compute the future projections using a multiplier factor (Equation18). As a result,
198 the monthly distribution's shape is preserved, and the delta shifts the values. This approach does not allow for
199 changes in extreme precipitation events due to its simple transfer function (Mendez et al. 2020). A multiplicative
200 correction is used for the precipitation correction equation.

*
201 Pcontrol ( d ) = Pobserved ( d ) (17)

éµ (P ( d )) ù
202 *
Pscenario ( d ) = Pobserved ( d ) × ê m scenario ú (18)
êë µm ( Pcontrol ( d ) ) úû

203

6
204 3.3 Future Projection of Precipitation:

205 Climate projections are based on the emission scenarios, so-called Radiative Concentration Pathways
206 (RCP), reflecting a shift in the radiative forcing at the atmosphere by 2100 compared to pre-industrial times (Van
207 Vuuren et al. 2011a). The four RCPs (RCP2.6, RCP4.5, RCP6, and RCP 8.5) are named based on the radiative
208 forcing change by 2100 (+ 2.6, + 4.5, + 6.0, and + 8.5 W/m2) respectively. In this study, CORDEX – REMO 2009
209 RCP 4.5 and RCP 8.5 were used for precipitation projection. The collected data were preprocessed using Climate
210 Data Operators (CDO 2019) and bias-correction using CMhyd software (Rathjens et al. 2016).

211 4.0 Results & Discussion:

212 4.1 Performance assessment of various bias correction methods on the control data:
213 After bias correction, the annual mean was calculated for both observed and bias-corrected control datasets.
214 Subsequently, a correlation was computed amongst them, and the same is shown in Figure 3. The figure shows that
215 the DC method adjusted control data has an absolute agreement with observed data. The agreement is since the
216 method equalizes the mean of both observed and adjusted control data. The LS and PT method yield the least bias
217 amongst the other methods, while the LOCI method underestimated the inter-annual variability in all locations.

218 On the contrary, the DM method has shown a varied result. It has overestimated the inter-annual variability
219 at many locations, while in a few places, it has underestimated and in some areas demonstrated a fair agreement.
220 Thus, all the bias correction methods can improve the spatial statistics of simulated mean precipitation is inferred.

221 The Root Mean Square Error (RMSE) and biases were calculated (Table 1). RMSE estimates the standard
222 deviation of the error distribution between the observed and adjusted control data. The DC method seems to yield no
223 error from the results, while the PT and LS methods with significantly less deviation of 0.059 and 0.063,
224 respectively. Conversely, the DM method has shown a moderate deviation (24.097) and the LOCI method with a
225 very high deviation (105.144). Thus it can be surmised that the PT bias correction method shows better performance
226 than others.
227 The bias indicates the difference between the adjusted control data and the observed data. Again the DC
228 method seems to be with no biases. LS method reduces the precipitation amount by 0.04 mm, and PT represents
229 negative bias, reducing the precipitation amount by 0.07 mm. The LOCI method also shows a strong negative bias,
230 which reduces the amount of precipitation by 97.76 mm. In contrast, the DM method yields a strong positive bias,
231 increasing the precipitation amount by 31.04 mm. Based on the bias calculation, it is presumed that the LS method is
232 better for bias correction, followed by the PT method.
233 Thus from the above analyses, the DC approach should be neglected for assessment because it has equalled
234 the observation with the current condition. Similarly, LOCI and DM methods were also to be avoided owing to their
235 over and underestimations. Both LS and PT methods have shown a good agreement with observed data almost in all
236 the analyses. Though LS accounts for the mean biases, it does not correct the biases in the variance. Thus it can be
237 surmised that the PT approach can be adapted to adjust the future projections in the study.
238
Bias Correction Methods RMSE Bias
Linear scaling 0.063 -0.044
Local intensity scaling 105.144 -97.758
Power transformation of precipitation 0.059 -0.073
Distribution mapping 24.097 31.039
Delta change 0.000 0.000
239 Table 1. showing the root mean square error and bias for various bias correction methods
240
241 4.2 Projections using bias-corrected precipitation in RCP scenario:
242 The biases in RCP 4.5 and 8.5 scenario data were corrected using the PT method. Subsequently, grouped
243 into three groups as the near range (2021 to 2050), mid-range (2051 to 2075), and far range (2076 to 2100).

244 4.2.1 Comparison between observed and adjusted control data (1976 – 2005):

7
245 Initially, the PT method's performance was evaluated by carrying out statistical analysis between the
246 observed and adjusted control data. The result shows that the model underestimates the intensity of the precipitation
247 until 1985 and then after exhibits both overestimation and good agreement with the observed data collectively
248 (Figure 4). Similarly, a mean positive deviation of 38.96% and a negative deviation of 23% are witnessed between
249 observed and adjusted control data (Table 2). Overall the model overestimates 14 times and underestimates 16 times
250 the intensity of precipitation in the 30-year time series. However, though the frequency of negative deviation is
251 higher, the intensity of precipitation it underestimates is lesser than the overestimation (Table 2).

252
253 Table 2. Comparison of adjusted scenario data with observed and bias corrected control data
254
255 4.2.2 The near range (2021 – 2050) of RCP 4.5 and RCP 8.5:

256 In general, precipitation shows an increase during the first decade and then after a gradual decrease in both
257 scenarios (Figure 4). The RCP 4.5 has shown higher precipitation during 2022, 2025, 2031 and 2033 while RCP 8.5
258 during 2027, 2039 and 2048. The calculated bias denotes that precipitation has increased by 80.99 mm in RCP 8.5
259 with an occurrence frequency of 17 than RCP 4.5 (Figure 5). Meanwhile, the comparison with observed data shows
260 increased precipitation in both RCP 4.5 (63.02%) and RCP 8.5 (80.85%) but with the same number of occurrences
261 (18). Similarly, an increase of 65.84% for the RCP 4.5 scenario and 66.02% for RCP 8.5 with the same number of
262 occurrences (20) is noticed with adjusted control data for 30 years time series (Table 2).

263 The spatial distribution of annual mean precipitation for both scenarios in the near range is represented in
264 figure 6 and figure 7. In RCP 4.5, most of the area receives precipitation of around 1150 mm. However, a decrease
265 in the west and increases towards the northeast is also noticed. The RCP 8.5 exhibits a similar spatial pattern;
266 however, most places have shown relatively increased precipitation of about ~1275 mm (Figure 6). From the wind
267 rose diagram (Figure 7), it is witnessed that Manjalar head and Lower anaicut area receives higher precipitation (~
268 1600 mm) whereas Grand anaicut and Thirukattupalli area with lower precipitation (~ 1000 mm).

269 Overall, higher precipitation of RCP 8.5 is attributed to radiative forcing. But the frequency of occurrence
270 seems to be similar for both RCP scenarios as far observed and adjusted control data is concerned. The same is
271 attributed to the effect of the adapted bias correction method.

272 4.2.3 The mid range (2051 – 2075) of RCP 4.5 and RCP 8.5:

273 It is found that RCP 4.5 receives higher precipitation during 2052, 2056 and 2065, while RCP 8.5 receives
274 more precipitation during the years 2053 and 2057. Based on the calculated bias, the amount of precipitation seems
275 to have increased by 99.19 mm in RCP 4.5 with an occurrence frequency of 12 compared to RCP 8.5 (Figure 5).
276 Similarly, the comparison with the observed data shows that RCP 4.5 yields an increased amount of precipitation
277 (54.35%) with a frequency rate of 17. In contrast, RCP 8.5 possesses 54.41% of increased precipitation with a

8
278 frequency rate of 15. The adjusted control data increases 91.73% for the RCP4.5 scenario, while the RCP 8.5 shows
279 an increase of 55.68% with the same number of occurrences (15) for the 25 years time series (Table 2).

280 The spatial distribution of annual mean precipitation for both scenarios in the mid range is represented in
281 figure 7 and figure 8. In RCP 4.5, the Adhiramapattinam, Lower anaicut, Manjalar head, and Pattukottai receive
282 more precipitation (~1400 mm), whereas Grand anaicut and Thirukattupalli receives low precipitation (~1000 mm).
283 While in RCP 8.5, the Lower anaicut, Manjalar head, Pattukottai, Papanasam, Kumbakonam, Vetticadu and
284 Thanjavur receives high precipitation (~1300 mm), and Thiruvaiyaaru, Grand anaicut and Thirukattupalli receives
285 low precipitation (~1100 mm).

286 Overall, it is found that the mid-range of RCP 4.5 receives higher precipitation than RCP 8.5. Thus, it is
287 inferred that the radiative forcing does not influence the intensity of the precipitation.

288 4.2.4 The far range (2076 – 2100) of RCP 4.5 and RCP 8.5:

289 The projected precipitation is overestimated by RCP 8.5 in 2082 with an unrealistic amount of ~5000 mm
290 (Figure 4). The precipitation intensity increases by 132.47 mm in RCP 8.5, with an occurrence frequency of 17 and
291 receiving higher precipitation during 2082 and 2093. Whereas in RCP 4.5, higher precipitation is witnessed during
292 2090 and 2092 (Figure 5). RCP 4.5 yields an increased precipitation of 56.73%, whereas 83.85% is for the RCP 8.5
293 scenario with the same frequency rate (18) against observed data. Concerning adjusted control data, an increase of
294 61.94% with an occurrence rate of 17 for RCP4.5 and for RCP 8.5, 81.16% with a frequency of 18 is witnessed for
295 25 years time series (Table 2).

296 The spatial distributions of annual mean precipitation for both scenarios in the far range are represented in
297 figure 7 and figure 9. Adhiramapattinam, Lower anaicut and Manjalar head shows high precipitation (~1400 mm),
298 whereas Grand anaicut and Thirukattupalli with low precipitation (~ 1000 mm) in RCP 4.5. on the contrary, RCP
299 8.5 has shown a more significant variation in the distribution pattern with high precipitation (~1750 mm) in Lower
300 anaicut, Manjalar head and Kumbakonam and low precipitation (~1100 mm) in Peravoorani, Thirukattipalli, and
301 Grand anaicut regions.
302
303 Overall in all the ranges, RCP 8.5 receives high precipitation when compared to RCP 4.5. The intensity of
304 precipitation increases towards the northeast and reduces towards the west. The Thirukattipalli and Grand anaicut
305 area located in the western part of the study exhibits very poor precipitation. The same might be attributed to the
306 effect of the northeast monsoon. Further, the research shows substantial interannual variabilities in both scenarios,
307 which might be due to the atmospheric circulation changes (Cha et al. 2016). The analogous spatial precipitation
308 distribution pattern in all ranges might be due to the horizontal resolution of the CORDEX data. An increase in
309 mean positive deviation (50%) with a high-frequency rate compared to mean negative deviation (30%) indicates that
310 the probability for flood is higher than the occurrence of drought in both RCPs.
311 According to RCP 4.5, the flood will occur during 2022, 2025, 2031, 2033, 2052, 2056, 2065, 2090 and
312 2092, whereas drought will occur during 2036, 2044, 2054, 2066, 2070, 2084 and 2095. According to RCP 8.5, the
313 flood will occur during 2027, 2039, 2048, 2053, 2057, 2082 and 2093, whereas drought will occur during 2023,
314 2030, 2034, 2055, 2059, 2062, 2065, 2071, 2091 and 2096. However, the effects of drought will be minimal because
315 of decreased precipitation and the absence of consecutive drought years.
316
317 5.0 Summary and Conclusion:

318 The choice of the bias correction algorithm plays a primary role in assessing the impacts of climate change.
319 This study provides an overview of the various bias correction methods and procedure for evaluating several
320 statistical parameters at the district scale. The improvement was achieved for the control data with all bias correction
321 methods with significant differences. Though all the methods are efficient in correcting the daily mean values, the
322 PT and DM methods can potentially fix the other statistical properties. Further, it is found that the Power
323 transformation method seems to be the best as far as this study area is concerned. The future climate projections of
324 the bias-corrected ensemble show considerable changes throughout the Century. Most of the years will receive
325 heavy precipitation; rather, some years will receive low and average precipitation than observed data. The spatial
326 distribution pattern indicates that the northeast monsoon will dominate over all the ranges and places.

9
327 This district is known for paddy cultivation. The projected precipitation will also influence the crop
328 selection, length of the growing period, cropping pattern, crop rotation, crop management practices, sowing period,
329 cropping area extent, agricultural production, and so on. In urban areas, the possibility for land use and land cover
330 changes, ground/river/surface water level changes, flood, soil erosion is high. Further, the study area is also known
331 for heritage and tourism. Thus the study has provided a piece of clear information on future precipitation to the
332 environmentalist, urban planners and policymakers of disaster management. Rainwater harvesting, recharging the
333 aquifers, afforestation, and redirecting the excess amount of water to the river through proper channels are plausible
334 suggestions to overcome excessive precipitation in the future.
335
336

337

338
339 Figure 1. Study area map showing Thanjavur district with block boundary and rain gauge stations
340
341

10
342
343 Figure 2. Methodology
344

345

346
347 Figure 3. Graphical representation of annual mean precipitation distribution of observed and bias
348 corrected control data for various bias correction methods with the resulted correlation coefficient
349

11
350
351 Figure 4. Plot showing time series annual mean precipitation (mm) for the observed, adjusted control and
352 the RCP scenarios in Thanjavur district.
353
354

355
356 Figure 5. The wind rose diagram showing the temporal distribution pattern of annual mean precipitation for RCP
357 4.5 and RCP 8.5 (2021 to 2100)
358
359

12
360
361 Figure 6. Map showing the spatial distribution pattern of annual mean precipitation for RCP 4.5 and RCP
362 8.5 in the near range period (2021-2050)
363

364

365
366 Figure 7. The wind rose diagram showing annual mean precipitation for both RCP scenarios in all three-time
367 ranges
368

13
369
370 Figure 8. Map showing the spatial distribution pattern of annual mean precipitation for RCP 4.5 and RCP
371 8.5 in the mid range period (2051-2075)
372
373

374
375 Figure 9. Map showing the spatial distribution pattern of annual mean precipitation for RCP 4.5 and RCP
376 8.5 in the far range period (2076-2100)

14
377
378 Declarations:
379 Conflict of Interest
380 The authors have no conflicts of interest to declare that are relevant to the content of this
381 manuscript.

382 Funding Statement

383 The authors did not receive support from any organization for the submitted work.

384 Author's Contribution

385 R. Selvakumar: Conceptualization, Methodology, Writing - review and editing, Supervision

386 S. Gunavathi: Analysis and investigation, Writing - original draft preparation

387 Availability of data and material

388 The datasets generated during and/or analysed during the current study are available from the
389 corresponding author on reasonable request.

390 Code availability

391 Not applicable
392
393 Ethics approval
394 Not applicable
395 Consent to participate
396 Not applicable
397 Consent for publication
398 Not applicable
399 References:
400
401 Alexander, L., Allen, S., & Bindoff, N. L. (2013) Working group I contribution to the IPCC fifth
402 assessment report climate change 2013: The physical science basis summary for policymakers (No. Bajados de
403 Internet/2013). OPCC.

404 Bennett, J. C. et al. (2011) Performance of Quantile-Quantile Bias-Correction for Use in

405 Hydroclimatological Projections. MODSIM 2011 - 19th International Congress on Modelling and Simulation -
406 Sustaining Our Future: Understanding and Living with Uncertainty (December), 2668–75.

407 Brent, R.P., (1971) An algorithm with guaranteed convergence for finding a zero of a function. Comput. J.
408 14 (4), 422-425. http://dx.doi.org/10.1093/comjnl/14.4.422.

409 Cha, D. H., Lee, D. K., Jin, C. S., Kim, G., Choi, Y., Suh, M. S., ... & Kang, H. S. (2016) Future changes in
410 summer precipitation in regional climate simulations over the Korean Peninsula forced by multi-RCP scenarios of
411 HadGEM2-AO. Asia-Pacific Journal of Atmospheric Sciences, 52(2), 139-149.

412 Chen, J., Brissette, F.P., Poulin, A., Leconte, R., (2011b) Overall uncertainty study of the hydrological
413 impacts of climate change for a Canadian watershed. Water

414 Chen J, Brissette FP, Chaumont D, Braun M. (2013) Finding appropriate bias correction methods in
415 downscaling precipitation for hydrologic impact studies over North America. Water Resour. Res. 49: 4187-4205.
416 https://doi.org/10.1002/wrcr.20331.

15
417 Christensen, J.H., Boberg, F., Christensen, O.B., Lucas-Picher, P., (2008) On the need for bias correction
418 of regional climate change projections of temperature and precipitation. Geophys. Res. Lett. 35 (20), L20709.
419 http://dx.doi.org/10.1029/2008GL035694.

420 Devi, Usha, M. S. Shekhar, and G. P. Singh. (2021) Correction of Mesoscale Model Daily Precipitation
421 Data over Northwestern Himalaya. Theoretical and Applied Climatology, 143(1–2), 51–60.

422 Dosio A, Paruolo P (2011) Bias correction of the ENSEMBLES highresolution climate change projections
423 for use by impact models: Evaluation on the present climate. J Geophys Res 116.
424 https://doi.org/10.1029/2011jd015934

425 Giorgi F, Mearns LO (1999) Introduction to special section: regional climate modeling revisited. J Geophys
426 Res-Atmos 104:6335-6352. https://doi.org/10.1029/98jd02072

427 Giorgi F, Jones C, Asrar GR (2009) Addressing climate information needs at the regional level: the
428 CORDEX framework World Meteorological Organization (WMO). Bulletin 58:175

429 Giorgi, F., & GAO, X. J. (2018) Regional earth system modeling: review and future directions.
430 Atmospheric and Oceanic Science Letters, 11(2), 189-197.

431 Gudmundsson, L., Bremnes, J. B., Haugen, J. E., & Engen-Skaugen, T. (2012) Downscaling RCM
432 precipitation to the station scale using statistical transformations-a comparison of methods. Hydrology and Earth
433 System Sciences, 16(9), 3383-3390.

434 Hanel M, Buishand TA (2011) Analysis of precipitation extremes in an ensemble of transient regional
435 climate model simulations for the Rhine basin. Clim Dyn 36:1135-1153. https://doi.org/10.1007/s00382-010-0822-2

436 Hall, A. (2014) Projecting regional change. Science, 346(6216), 1461-1462.

437 https://doi.org/10.1126/science.aaa0629yar

438 Ines AVM, Hansen JW. (2006) Bias correction of daily GCM rainfall for crop simulation studies. Agric.
439 For. Meteorol. 138: 44-53. https://doi .org/10.1016/j.agrformet.2006.03.009.

440 Johnson, C. R., Banks, S. C., Barrett, N. S., Cazassus, F., Dunstan, P. K., Edgar, G. J., ... & Taw, N.
441 (2011)Climate change cascades: Shifts in oceanography, species' ranges and subtidal marine community dynamics
442 in eastern Tasmania. Journal of Experimental Marine Biology and Ecology, 400(1-2), 17-32.

443 Kim, G., Cha, D. H., Lee, G., Park, C., Jin, C. S., Lee, D. K., ... & Kim, J.(2020) Projection of future
444 precipitation change over South Korea by regional climate models and bias correction methods.

445 Kumar, P., Wiltshire, A., Mathison, C., Asharaf, S., Ahrens, B., Lucas-Picher, P., ... & Jacob, D. (2013)
446 Downscaled climate change projections with uncertainty assessment over India using a high resolution multi-model
447 approach. Science of the Total Environment, 468, S18-S30.

448 Lafon, Thomas, Simon Dadson, Gwen Buys, and Christel Prudhomme. (2013) Bias Correction of Daily
449 Precipitation Simulated by a Regional Climate Model: A Comparison of Methods. International Journal of
450 Climatology, 33(6), 1367–81.

451 Mendez, M., Maathuis, B., Hein-Griggs, D., & Alvarado-Gamboa, L. F. (2020) Performance Evaluation of
452 Bias Correction Methods for Climate Change Monthly Precipitation Projections over Costa Rica. Water, 12(2), 482.

16
453 Middelkoop, H., Daamen, K., Gellens, D., Grabs, W., Kwadijk, J.C. J., Lang, H., Parmet, B., Schadler, B.,
454 Schulla, J., and Wilke,K. (2001) Impact of climate change on hydrological regimes and waterresources
455 management in the Rhine Basin. Clim. Change, 49, 105–128.

456 Monhart S, Spirig C, Bhend J, Bogner K, Schar C, Liniger MA (2018) Skill of subseasonal forecasts in
457 Europe: effect of bias correction and downscaling using surface observations. J Geophys Res-Atmos 123:7999-
458 8016. https://doi.org/10.1029/2017jd027923

459 Noguer, M., Jones, R., & Murphy, J. (1998) Sources of systematic errors in the climatology of a regional
460 climate model over Europe. Climate Dynamics, 14(10), 691-712.

461 N'Tcha M'Po, Yèkambèssoun. (2016) Comparison of Daily Precipitation Bias Correction Methods Based
462 on Four Regional Climate Model Outputs in Ouémé Basin, Benin. Hydrology, 4(6), 58.

463 Piani, C., J. O. Haerter, and E. Coppola. (2010) Statistical Bias Correction for Daily Precipitation in
464 Regional Climate Models over Europe. Theoretical and Applied Climatology, 99(1–2), 187–92.

465 PontoppidanM, Kolstad EW, Sobolowski S, King MP (2018) Improving the reliability and added value of
466 dynamical downscaling via correction of large-scale errors: a Norwegian perspective. J Geophys Res-Atmos
467 123:11,875-811,888. https://doi.org/10.1029/2018jd028372

468 Prusty, R. M., Das, A., & Patra, K. C. (2018) Climate change impact assessment under CORDEX South-
469 Asia RCM scenarios on water resources of the Brahmani and Baitarini River Basin, India.

470 Rathjens, H., Bieger, K., Srinivasan, R., Chaubey, I., & Arnold, J. G. (2016) CMhyd User Manual.

471 Räty, Olle, Jouni Räisänen, and Jussi S. Ylhäisi. (2014) Evaluation of Delta Change and Bias Correction
472 Methods for Future Daily Precipitation: Intermodel Cross-Validation Using ENSEMBLES Simulations. Climate
473 Dynamics, 42(9–10), 2287–2303.

474 Rauscher, S. A., Coppola, E., Piani, C., & Giorgi, F. (2010) Resolution effects on regional climate model
475 simulations of seasonal precipitation over Europe. Climate dynamics, 35(4), 685-711

476 Schmidli, J., Frei, C., & Vidale, P. L. (2006) Downscaling from GCM precipitation: a benchmark for
477 dynamical and statistical downscaling methods. International Journal of Climatology: A Journal of the Royal
478 Meteorological Society, 26(5), 679-689.

479 Schulzweida, Uwe. (2019) CDO User Guide (Version 1.9.8). http://doi.org/10.5281/zenodo.3539275).

480 Shrestha, Durga Lal, David E. Robertson, James C. Bennett, and Q. J. Wang. (2015) Improving
481 Precipitation Forecasts by Generating Ensembles through Postprocessing. Monthly Weather Review, 143(9), 3642–
482 63.

483 Terink, W, R T W L Hurkmans, P J J F Torfs, and R Uijlenhoet (2009) Bias Correction of Temperature and
484 Precipitation Data for Regional Climate Model Application to the Rhine Basin. Hydrology and Earth System
485 Sciences Discussions, 6(4), 5377–5413.

486 Teutschbein C, Seibert J (2012) Bias correction of regional climate model simulations for hydrological
487 climate-change impact studies: review and evaluation of different methods. J Hydrol 456:12-29. https://
488 doi.org/10.1016/j.jhydrol.2012.05.052

489 Teutschbein, C., & Seibert, J. (2013) Is bias correction of regional climate model (RCM) simulations
490 possible for non-stationary conditions?. Hydrology and Earth System Sciences, 17(12), 5061-5077.

17
491 Themeßl, M. J., Gobiet, A., & Heinrich, G. (2012) Empirical-statistical downscaling and error correction of
492 regional climate models and its impact on the climate change signal. Climatic Change, 112(2), 449-468.

493 Van Vuuren DP et al (2011) The representative concentration pathways: an overview. Clim Change, 109:5-
494 31. https://doi.org/10.1007/ s10584-011-0148-z.

495 Veijalainen N, Lotsari E, Alho P, Vehviläinen B, Käyhkö J (2010) National scale assessment of climate
496 change impacts on flooding in Finland. J Hydrol 391:333-350.

497

18
Figures

Figure 1

Study area map showing Thanjavur district with block boundary and rain gauge stations. Note: The
designations employed and the presentation of the material on this map do not imply the expression of
any opinion whatsoever on the part of Research Square concerning the legal status of any country,
territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This
map has been provided by the authors.
Figure 2

Methodology
Figure 3

Graphical representation of annual mean precipitation distribution of observed and bias corrected control
data for various bias correction methods with the resulted correlation coe cient

Figure 4

Plot showing time series annual mean precipitation (mm) for the observed, adjusted control and the RCP
scenarios in Thanjavur district.
Figure 5

The wind rose diagram showing the temporal distribution pattern of annual mean precipitation for RCP
4.5 and RCP 8.5 (2021 to 2100)
Figure 6

Map showing the spatial distribution pattern of annual mean precipitation for RCP 4.5 and RCP 8.5 in the
near range period (2021-2050). Note: The designations employed and the presentation of the material on
this map do not imply the expression of any opinion whatsoever on the part of Research Square
concerning the legal status of any country, territory, city or area or of its authorities, or concerning the
delimitation of its frontiers or boundaries. This map has been provided by the authors.
Figure 7

The wind rose diagram showing annual mean precipitation for both RCP scenarios in all three-time
ranges
Figure 8

Map showing the spatial distribution pattern of annual mean precipitation for RCP 4.5 and RCP 8.5 in the
mid range period (2051-2075). Note: The designations employed and the presentation of the material on
this map do not imply the expression of any opinion whatsoever on the part of Research Square
concerning the legal status of any country, territory, city or area or of its authorities, or concerning the
delimitation of its frontiers or boundaries. This map has been provided by the authors.
Figure 9

Map showing the spatial distribution pattern of annual mean precipitation for RCP 4.5 and RCP 8.5 in the
far range period (2076-2100). Note: The designations employed and the presentation of the material on
this map do not imply the expression of any opinion whatsoever on the part of Research Square
concerning the legal status of any country, territory, city or area or of its authorities, or concerning the
delimitation of its frontiers or boundaries. This map has been provided by the authors.